Skip to content

Latest commit



102 lines (67 loc) · 3.22 KB


File metadata and controls

102 lines (67 loc) · 3.22 KB

Submitting Applications


The submission API is experimental and may change between versions

Sometimes you have Dask Application you want to deploy completely on YARN, without having a corresponding process running on an edge node. This may come up with production applications deployed automatically, or long running jobs you don't want to consume edge node resources.

To handle these cases, dask-yarn provides a cli that can be used to submit applications to be run on the YARN cluster asynchronously. There are three commands that may be useful here:

  • dask-yarn submit: submit an application to the YARN cluster
  • dask-yarn status: check on the status of an application
  • dask-yarn kill: kill a running application

Submitting an Application


To prepare an application to be submitted using dask-yarn submit, you need to change the creation of your YarnCluster from using the constructor to using YarnCluster.from_current.

# Replace this
cluster = YarnCluster(...)

# with this
cluster = YarnCluster.from_current()

This is because the script won't be run until the cluster is already created -at that point configuration passed to the YarnCluster constructor won't be useful. Cluster configuration is instead passed via the dask-yarn submit CLI (note that as before, the cluster can be scaled dynamically after creation).

# Submit `` to run on a dask cluster with 8 workers,
# each with 2 cores and 4 GiB
$ dask-yarn submit \
  --environment my_env.tar.gz \
  --worker-count 8 \
  --worker-vcores 2 \
  --worker-memory 4GiB \


This outputs a YARN Application ID, which can be used with other YARN tools.

Checking Application Status

Submitted application status can be checked using the YARN Web UI, or programmatically using dask-yarn status. This command takes one parameter -the application id.

$ dask-yarn status application_1538148161343_0051
application_1538148161343_0051    dask    RUNNING     UNDEFINED    9             17        33792     6m

Killing a Running Application

Submitted applications normally run until completion. If you need to terminate one before then, you can use the dask-yarn kill command. This command takes one parameter - the application id.

$ dask-yarn kill application_1538148161343_0051

Accessing the Application Logs

Application logs can be retrieved a few ways:

  • The logs of running applications can be viewed using the Skein Web UI (dask-yarn is built using Skein).
  • The logs of completed applications can be viewed using the yarn logs command.

    $ yarn logs -applicationId application_1538148161343_0051